Search

This study examines the neural activities of participants undergoing vibro-motor reprocessing therapy (VRT) while experiencing motion sickness. We evaluated the efficacy of vibro-motor reprocessing therapy, a novel therapeutic technique based on eye movement desensitization and reprocessing (EMDR), in reducing motion sickness. Based on visually induced motion sickness in two sets of performed sessions, eight participants were exposed to VRT stimulation in a VRT/non-VRT setting. Simultaneously, brain activity changes were recorded using electroencephalography (EEG) at baseline and during stimulus exposure, and comparisons made across the VRT/non-VRT conditions. A significant reduction in the alpha (8-12 Hz) spectral power was observed in the frontal and occipital locations, consistent across all participants. Furthermore, significant reductions were also found in the frontal and occipital delta (0.5-4 Hz) and theta (4-8 Hz) spectral power frequency bands between non-VRT and VRT conditions (p0.05). Our results offer novel insights for a potential nonpharmacological treatment and attenuation of motion sickness. Furthermore, symptoms can be observed, and alleviated, in real-time using the reported techniques. Clinical relevance - Instead of using drugs to treat motion sickness, patients could safely use this VRT technique.

This paper evaluates a range of deep learning frameworks for detecting respiratory anomalies from input audio. Audio recordings of respiratory cycles collected from patients are transformed into time-frequency spectrograms to serve as front-end two-dimensional features. Cropped spectrogram segments are then used to train a range of back-end deep learning networks to classify respiratory cycles into predefined medically-relevant categories. A set of those trained high-performance deep learning frameworks are then fused to obtain the best score. Our experiments on the ICBHI benchmark dataset achieve the highest ICBHI score to date of 57.3%. This is derived from a late fusion of inception based and transfer learning based deep learning frameworks, easily outperforming other state-of-the-art systems. Clinical relevance--- Respiratory disease, wheeze, crackle, inception, convolutional neural network, transfer learning.

End-to-end attention-based modeling is increasingly popular for tackling sequence-to-sequence mapping tasks. Traditional attention mechanisms utilize prior input information to derive attention, which then conditions the output. However, we believe that knowledge of posterior output information may convey some advantage when modeling attention. A recent technique proposed for machine translation called the posterior attention model (PAM) demonstrates that posterior output information can be used in that way for machine translation. This paper explores the use of posterior information for attention modeling in an automatic speech recognition (ASR) task. We demonstrate that direct application of PAM to ASR is unsatisfactory, due to two deficiencies; Firstly, PAM adopts attention based weighted single-frame output prediction by assuming a single focused attention variable, whereas wider contextual information from acoustic frames is important for output prediction in ASR. Secondly, in addition to the well-known exposure bias problem, PAM introduces additional mismatches in attention training and inference calculations. We present extensive experiments combining a number of alternative approaches to solving these problems, leading to a high performance technique which we call extended PAM (EPAM). To counter the first deficiency, EPAM modifies the encoder to introduce additional context information for output prediction. The second deficiency is overcome in EPAM through a two part solution of a mismatch penalty term and an alternate learning strategy. The former applies a divergence-based loss to correct the mismatch bias distribution, while the latter employs a novel update strategy which relies on introducing iterative inference steps alongside each training step. In experiments with both WSJ-80hrs and Switchboard-300hrs datasets we found significant performance gains. For example, the full EPAM system model achieved a word error rate (WER) of 10.6% on the WSJ eval92 test set, compared to 11.6% for traditional prior-attention modeling. Meanwhile, on the Switchboard eval2000 test set, we achieved 16.3% WER compared to the traditional method WER of 17.3%.

Passive detection of footsteps in domestic settings can allow the development of assistive technologies that can monitor mobility patterns of older adults in their home environment. Acoustic footstep detection is a promising approach for non-intrusive detection of footsteps. So far there has been limited work in developing robust acoustic footstep detection systems that can operate in noisy home environments. In this paper, we propose a novel application of the Attention based Recurrent Deep Neural Network to detect human footsteps in noisy overlapping audio streams. The model is trained on synthetic data which simulates the acoustic scene in a home environment. To evaluate performance, we reproduced two footstep detection models from literature and compared them using the newly developed Polyphonic Sound Detection Scores (PSDS). Our model achieved the highest PSDS and is close to the highest score achieved by generic indoor AED models in DCASE. The proposed system is designed to both detect and track footsteps within a home setting, and to enhance state-of-the-art digital health-care solutions for empowering older adults to live autonomously in their own homes.

This paper presents an inception-based deep neural network for detecting lung diseases using respiratory sound input. Recordings of respiratory sound collected from patients are first transformed into spectrograms where both spectral and temporal information are well represented, in a process referred to as front-end feature extraction. These spectrograms are then fed into the proposed network, in a process referred to as back-end classification, for detecting whether patients suffer from lung-related diseases. Our experiments, conducted over the ICBHI benchmark metadataset of respiratory sound, achieve competitive ICBHI scores of 0.53/0.45 and 0.87/0.85 regarding respiratory anomaly and disease detection, respectively.

In this paper, we present an effective end-to-end deep embedding learning method based on Dense-Residual networks, which combine the advantages of a densely connected convolutional network (DenseNet) and a residual network (ResNet), for speaker verification (SV). Unlike a model ensemble strategy which merges the results of multiple systems, the proposed Dense-Residual networks perform feature fusion on every basic DenseR building block. Specifically, two types of DenseR blocks are designed. A sequential-DenseR block is constructed by densely connecting stacked basic units in a residual block of ResNet. A parallel-DenseR comprises split and concatenation operations on residual and dense components via corresponding skip connections. These building blocks are stacked into deep networks to exploit the complementary information with different receptive field sizes and growth rates. Extensive experiments have been conducted on the VoxCeleb1 dataset to evaluate the proposed methods. The SV performance achieved by the proposed Dense-Residual networks is shown to outperform corresponding ResNet, DenseNet or fusions of them, with similar model complexity, by a significant margin.

This paper presents an improved mean teacher (MT) based method for large-scale weakly labeled semi-supervised sound event detection (SED), by focusing on learning a better student model. Two main improvements are proposed based on the authors’ previous perturbation based MT method. Firstly, an event-aware module is de-signed to allow multiple branches with different kernel sizes to be fused via an attention mechanism. By inserting this module after the convolutional layer, each neuron can adaptively adjust its receptive field to suit different sound events. Secondly, instead of using the teacher model to provide a consistency cost term, we propose using a stochastic inference of unlabeled examples to generate high quality pseudo-targets by averaging multiple predictions from the perturbed student model. MixUp of both labeled and unlabeled data is further exploited to improve the effectiveness of student model. Finally, the teacher model can be obtained via exponential moving average (EMA) of the student model, which generates final predictions for SED during inference. Experiments on the DCASE2018 task4 dataset demonstrate the ability of the proposed method. Specifically, an F1-score of 42.1% is achieved, significantly outperforming the 32.4% achieved by the winning system, or the 39.3% by the previous perturbation based method.

In order to reduce motion sickness (MS), a novel vibro-motor reprocessing therapy (VRT), based on an Eye Movement Desensitization and Reprocessing (EMDR) technique is investigated. To the best of the authors knowledge, this is the first time that reprocessing therapy has been evaluated for alleviating MS. Experimentally, MS was induced using visual stimulus of motion videos. Subjective MS was then recorded at baseline for both VRT and non-VRT stimulation conditions, and after each condition, evaluated using a Motion Sickness Assessment Questionnaire (MSAQ). MSAQ scores were compared for both conditions in eight test subjects, with a significant and clear reduction in motion sickness symptoms revealed when applying VRT stimulation. While the subject pool is small, this pilot study indicates that the proposed approach has potential for future exploration in terms of non-pharmacological treatment and management of MS.

Modern audiometry is largely a behavioural task, with the pure-tone audiogram (PTA) being the gold standard for evaluating frequency-specific hearing thresholds in adults. The nature of behavioural audiometry makes estimating accurate hearing thresholds difficult in infants and people with disabilities, where following instructions or interacting with the test may be difficult or impossible. We propose a method in which Auditory Evoked Potentials (AEPs) are used as an alternative to behavioural audiometry for detecting frequency-specific thresholds. Specifically, P300 responses elicited by the tones of a PTA are automatically detected from electroencephalogram (EEG) data, to evaluate hearing acuity. To assess the effectiveness of this method, we created a dataset of EEG recordings from participants presented with a series of pure tones at 6 different frequencies with steadily decreasing volumes, during a PTA test. This dataset was used to train a support vector machine (SVM) to identify when a participant was played a tone, whether they perceived it or not using their EEG. Results demonstrate that detecting hearing events can be very accurate for participants for whom the classifier has been trained apriori. However, accuracy drops significantly for unseen participants - when the classifier has not been trained on any prior data from a given participant before classifying their EEG. However, by establishing that AEP response-based audiometry is viable for detecting tones, future work will explore the ability of more powerful deep neural networks to accurately estimate for unseen participants.

This study explores auricular vagus nerve stimulation (aVNS) within the context of stress. Five healthy subjects underwent a pulsed mode aVNS (with a frequency of 25 Hz and pulse width of $200\ \mu\mathrm{s}$ ) using a custom made current stimulation device. The device triggered the auricular vagus nerve branch through the tragus for 15 minutes, with prefrontal EEG data collected pre and post aVNS (each for 5 minutes) to capture stress indicators. Given that the frontal location is prone to noise, EEG signal pre-processing through independent component analysis (ICA) was used to reduce eye movements and other artifacts. Alpha (8-14 Hz) asymmetry in prefrontal location in both hemispheres was computed from locations Fp1 and Fp2 for one second segments of EEG; these asymmetry values were normalised by subtracting the post aVNS (or placebo) from pre aVNS (placebo). The normalised alpha frontal asymmetries were assessed for statistical significance using T-test. Results showed increased aVNS alpha prefrontal asymmetry compared to placebo, indicating reduced stress. This was statistically significant ( $\mathrm{p} ) for all five subjects. It is therefore concluded that the pulsed mode aVNS stimulation at the tragus may represent a potential approach for stress relief.